Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76597.42746771034,
-      "required_pages_per_second_upper_estimate": 0.005299187762734413
+      "required_token_positions_per_second": 77202.60894777653,
+      "required_pages_per_second_upper_estimate": 0.0053410556217398895
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9574.678433463792,
-      "required_pages_per_second_upper_estimate": 0.004008717918069434
+      "required_token_positions_per_second": 9650.326118472067,
+      "required_pages_per_second_upper_estimate": 0.004040390024079101
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19149.356866927585,
-      "required_pages_per_second_upper_estimate": 0.003988306032880488
+      "required_token_positions_per_second": 19300.652236944134,
+      "required_pages_per_second_upper_estimate": 0.004019816868527717
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28724.035300391377,
-      "required_pages_per_second_upper_estimate": 0.0017219620571960815
+      "required_token_positions_per_second": 28950.9783554162,
+      "required_pages_per_second_upper_estimate": 0.0017355669468228387
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28724.035300391377,
-      "required_pages_per_second_upper_estimate": 0.002233167400926555
+      "required_token_positions_per_second": 28950.9783554162,
+      "required_pages_per_second_upper_estimate": 0.0022508112252377305
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28724.035300391377,
-      "required_pages_per_second_upper_estimate": 0.3684410833672141
+      "required_token_positions_per_second": 28950.9783554162,
+      "required_pages_per_second_upper_estimate": 0.37135206520460484
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0793128045701792,
-    "producer_p95_seconds": 0.28144279299885966,
-    "gpu_update_seconds": 0.9625388532934746,
-    "producer_batches_per_second": 12.608304616377035,
-    "gpu_batches_per_second": 1.038919100853276,
-    "producer_headroom": 12.135983067422373,
+    "producer_batch_seconds": 0.23279623091434587,
+    "producer_p95_seconds": 0.28698391799844103,
+    "gpu_update_seconds": 0.9549936330503167,
+    "producer_batches_per_second": 4.295602192837633,
+    "gpu_batches_per_second": 1.0471274000078197,
+    "producer_headroom": 4.102272744276918,
     "sustainable": true,
-    "buffer_seconds": 60.6399477574889,
+    "buffer_seconds": 60.164598882169955,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 364340.5899952675,
-    "upload_bytes_per_second": 5002833.013420171,
+    "generation_bytes_per_second": 394307.6412913867,
+    "upload_bytes_per_second": 4986776.241108664,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.17900919812698,
+    "estimated_full_shards_per_hour": 84.60924080902291,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.2006459389995143,
-      "bytes": 7651087,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.3339825349976309,
       "bytes": 10020438,
@@ -562,6 +556,12 @@
       "bytes": 11211496,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 1.1181462930035195,
+      "bytes": 5980396,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -316,3 +316,4 @@
 {"step": 716600, "loss": 0.5833146572113037, "lr": 8e-05, "grad_norm": 0.22927128309571412, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104890590349, "tokens_per_second": 153805.43548895695, "data_wait_s": 1.6591002349741757e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 716700, "loss": 0.5062534809112549, "lr": 8e-05, "grad_norm": 0.20410455130525226, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104905129655, "tokens_per_second": 151253.71473505115, "data_wait_s": 1.3720004062633961e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 716800, "loss": 0.55482417345047, "lr": 8e-05, "grad_norm": 0.21592001114608955, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104919723214, "tokens_per_second": 151721.58774141202, "data_wait_s": 1.7788996046874672e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 716900, "loss": 0.5516698360443115, "lr": 8e-05, "grad_norm": 0.21892667708024693, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104934179270, "tokens_per_second": 150519.78573794657, "data_wait_s": 1.4702003682032228e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:968ccad0fd4e7ce9f01c6160ca9dd3a9d4aa3e543feb06ea10bd4b009fb8aac7
-size 678387
+oid sha256:7090b2bbb8788f3a1c5a7e6be04aa6b6f64f58a7a2635166f7bad1fe04dfaf35
+size 680505